In-Context Multiple Instance Learning

Hugging Face Daily Papers Papers

Summary

This paper proposes a Perceiver-style architecture pretrained on synthetic bag-structured data to enable efficient, task-adaptive classification from few labeled examples in multiple instance learning, outperforming supervised baselines across twelve benchmarks.

Multiple Instance Learning (MIL) addresses problems where supervision is available at the level of bags of instances and has been successfully applied in fields ranging from computational pathology to satellite imagery. Nevertheless, existing algorithms struggle in the low-label regime that characterizes many real-world applications. Flexible models overfit and rigid ones fail to adapt to the task at hand. We show that pretraining an in-context learner with a Perceiver-style architecture on synthetic data yields a model that can solve new tasks from a handful of labeled bags. At inference time, classification happens in a single forward pass and requires no gradient updates. We propose and investigate different synthetic data generators for bag-structured data and find that they capture complementary inductive biases. A model pretrained on a mixture of these generators inherits their per-task strengths and achieves the best average performance across twelve MIL benchmarks, outperforming supervised baselines that require task-specific training.
Original Article
View Cached Full Text

Cached at: 06/10/26, 05:46 PM

Paper page - In-Context Multiple Instance Learning

Source: https://huggingface.co/papers/2606.06458

Abstract

Pretraining a Perceiver-style architecture on synthetic bag-structured data enables efficient, task-adaptive classification from few labeled examples in multiple instance learning scenarios.

Multiple Instance Learning(MIL) addresses problems where supervision is available at the level of bags of instances and has been successfully applied in fields ranging from computational pathology to satellite imagery. Nevertheless, existing algorithms struggle in the low-label regime that characterizes many real-world applications. Flexible models overfit and rigid ones fail to adapt to the task at hand. We show thatpretrainingan in-context learner with aPerceiver-style architectureonsynthetic datayields a model that can solve new tasks from a handful of labeled bags. At inference time, classification happens in a single forward pass and requires no gradient updates. We propose and investigate differentsynthetic datagenerators forbag-structured dataand find that they capture complementary inductive biases. A model pretrained on a mixture of these generators inherits their per-task strengths and achieves the best average performance across twelve MIL benchmarks, outperforming supervised baselines that require task-specific training.

View arXiv pageView PDFGitHub1Add to collection

Get this paper in your agent:

hf papers read 2606\.06458

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2606.06458 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2606.06458 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2606.06458 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

MAG: MAnifold Guided Semi-Supervised Multi-modal In-Context Learning

arXiv cs.LG

This paper introduces MAG, a manifold-guided framework for semi-supervised multi-modal in-context demonstration selection, leveraging unlabeled data to improve few-shot ICL for MLLMs. Experiments on eight benchmarks show consistent gains in label-scarce regimes.

In-Context Learning Operates as Concept Subspace Learning

arXiv cs.LG

This paper proposes that in-context learning in LLMs operates through low-dimensional concept subspaces, where task-relevant information concentrates in a small fraction of the representation space, supported by experiments on Llama-3-8B and Qwen2.5-7B.